SARS-CoV-2 concentration in wastewater consistently predicts trends in COVID-19 case counts by at least two days across multiple WWTP scales

Abstract

Wastewater surveillance of SARS-CoV-2 has proven instrumental in mitigating the spread of COVID-19 by providing an economical and equitable approach to disease surveillance. Here, we analyze the correlation of SARS-CoV-2 RNA in influents of seven wastewater plants (WWTPs) across the state of South Carolina with corresponding daily case counts to determine whether underlying characteristics of WWTPs and sewershed populations predict stronger correlations. The populations served by these WWTPs have varying social vulnerability and represent 24% of the South Carolina population. The study spanned 15 months from April 19, 2020, to July 1, 2021, which includes the administration of the first COVID-19 vaccines. SARS-CoV-2 RNA concentrations were measured by either reverse transcription quantitative PCR (RT-qPCR) or droplet digital PCR (RT-ddPCR). Although populations served and average flow rate varied across WWTPs, the strongest correlation was identified for six of the seven WWTPs when daily case counts were lagged two days after the measured SARS-CoV-2 RNA concentration in wastewater. The weakest correlation was found for WWTP 6, which had the lowest ratio of population served to average flow rate, indicating that the SARS-CoV-2 signal was too dilute for a robust correlation. Smoothing daily case counts by a 7-day moving average improved correlation strength between case counts and SARS-CoV-2 RNA concentration in wastewater while dampening the effect of lag-time optimization. Correlation strength between cases and SARS-CoV-2 RNA was compared for cases determined at the ZIP-code and sewershed levels. The strength of correlations using ZIP-code-level versus sewershed-level cases were not statistically different across WWTPs. Results indicate that wastewater surveillance, even without normalization to fecal indicators, is a strong predictor of clinical cases by at least two days, especially when SARS-CoV-2 RNA is measured using RT-ddPCR. Furthermore, the ratio of population served to flow rate may be a useful metric to assess whether a WWTP is suitable for a surveillance program.

Graphical abstract

1. Introduction

The ongoing COVID-19 pandemic, instigated by the severe acute respiratory syndrome coronavirus 2, or SARS-CoV-2, necessitated the development of techniques to monitor viral spread and evolution of new variants across the world. Since SARS-CoV-2 is shed in feces (Zhang et al., 2020) and trends in concentration often precede clinical data (Wu et al., 2022), wastewater surveillance has emerged at the forefront of these techniques, complementing clinical data, and providing an early warning of the presence of variants of concern (Rios et al., 2021; Swift et al., 2021; Vo et al., 2022). Surveillance efforts take place at multiple scales: building or facility (Gibas et al., 2021; Scott et al., 2021; Sellers et al., 2022; Vo et al., 2022), neighborhood (Ahmed et al., 2021; Haak et al., 2022; Rios et al., 2021), and city or wastewater treatment plant (Gonzalez et al., 2020; Smyth et al., 2022; Swift et al., 2021). The U.S. Centers for Disease Control and Prevention established a national wastewater surveillance system (NWSS) to track SARS-CoV-2 levels in wastewater nationwide to better inform community decisions to minimize the spread of COVID-19 (Kirby et al., 2021). Since wastewater surveillance is less expensive compared to monitoring the same number of individuals by clinical testing, it can be implemented in low to middle income countries (Shrestha et al., 2021), which furthers a more equitable response to global health crises such as pandemics.

Many studies have reported both qualitative and quantitative agreement between trends in SARS-CoV-2 RNA and trends in COVID-19 case counts (Gonzalez et al., 2020; Medema et al., 2020; Weidhaas et al., 2021; Wu et al., 2022; Wurtzer et al., 2022, 2020). The majority of tracking studies measured various SARS-CoV-2 targets by reverse transcription quantitative PCR (RT-qPCR), with fewer studies measuring SARS-CoV-2 targets using absolute quantification by reverse transcription droplet digital PCR (RT-ddPCR), such as the study by Gonzalez and colleagues (Gonzalez et al., 2020). RT-ddPCR has been demonstrated as more sensitive than RT-qPCR for the detection of SARS-CoV-2 from rhino-pharyngeal samples (Falzone et al., 2020; Suo et al., 2020), although RT-ddPCR is typically higher cost (Cervantes-Avilés et al., 2021). Higher sensitivity and precision in the detection of SARS-CoV-2 genetic targets from wastewater was demonstrated for RT-ddPCR compared to RT-qPCR (Flood et al., 2021). The approaches used to correlate community infection rates with SARS-CoV-2 concentration in wastewater have also differed, with many alternative gene targets used, such as the nucleocapsid and envelope genes. One of the most notable differences in these correlations is the lag time between case counts and measured SARS-CoV-2 signal in wastewater. An early tracking study reported that SARS-CoV-2 trends in wastewater foreshadowed trends in case counts by 4–10 days (Wu et al., 2022). Weidhaas and coworkers found significant correlations between viral load and weekly case rates for two wastewater treatment plants (WWTPs), but noted that an additional WWTP in their study required case rates to lag one week behind SARS-CoV-2 concentration in wastewater to identify a significant correlation (Weidhaas et al., 2021). During the first month of the pandemic in the Netherlands, Medema and colleagues reported a significant correlation between SARS-CoV-2 concentration in WWTP influent and cumulative city-level COVID-19 prevalence (Medema et al., 2020). As the pandemic progressed, Wurtzer and colleagues calculated the correlation between the log-transformed COVID-19 incidence rate and the wastewater indicator, which adjusts for flow rates and uses a Kalman smoother mathematical model followed by a normalizer to account for differences between WWTPs (Wurtzer et al., 2022). They identified an average 3-day lag between the wastewater signal and the 7-day COVID-19 incidence. In another study, Melvin and colleagues developed Melvin's index in order to normalize between different primer/probe sets and reduce bias when using RT-qPCR measurements near the limit of detection (Melvin et al., 2021). Their correlations between Melvin's index and daily case counts were optimized with a lag time of up to 23 days for some regions. These results suggest that different approaches will influence the correlation strength as well as the optimal lag time between COVID-19 case counts and SARS-CoV-2 signal in wastewater. Beyond linear correlations of case counts to SARS-CoV-2 concentration in wastewater, more advanced approaches to modeling in wastewater epidemiology studies of SARS-CoV-2 have been developed, such as those incorporating artificial neural networks (Jiang et al., 2022) or mass rates of SARS-CoV-2 in wastewater (McMahan et al., 2021). However, it is not fully understood how WWTP and community characteristics such as flow rate, population served, or infection rate also affect the maximum correlation strength and optimal lag.

A coordinated statewide wastewater surveillance effort was led in South Carolina spanning geographically distinct cities. In total, seven WWTPs were monitored biweekly for SARS-CoV-2 RNA concentration in the influent over a period spanning 15 months from April 19, 2020, to July 1, 2021, making this study one of the longest wastewater surveillance efforts in the United States during the COVID-19 pandemic. The total population served by all WWTPs was 1.2 million individuals, representing approximately one quarter of the entire population of South Carolina. The extended duration of the study and range of WWTPs included facilitated the testing of multiple hypotheses. Namely, we tested whether the strength of the correlation between SARS-CoV-2 RNA concentration in wastewater and daily clinical case count was related to the infection load (number of infected individuals per population served) and the ratio of the number of individuals served to the average flow rate. We hypothesized that weaker correlations would be found between viral RNA concentrations and case counts at WWTPs that served communities with low infection loads and high flow rates. Furthermore, we tested whether using a seven-day moving average of daily case counts improved the correlation with SARS-CoV-2 concentration in wastewater as well as the optimal lag time of case counts after wastewater collection.

2. Materials and methods

2.1. Wastewater sampling and sewershed population and COVID-19 case count estimates

In partnership with the South Carolina Department of Health and Environmental Control (SCDHEC), statewide wastewater samples were collected from April 19, 2020, to July 1, 2021. In total, seven WWTPs were sampled (Fig. 1 ). Characteristics of each WWTP included in the study, such as average influent flow rate and population served, are included in Table 1 and Supplementary Table S1. The populations served by the WWTPs varied from 22,266 to 363,714 individuals served and associated flow rates ranged from approximately two million gallons per day (MGD) to 40 MGD. The population was approximated for each sewershed from ZIP-code-level 2019 U.S. Census estimate. ZIP-codes only partially within the sewershed boundary were counted in their entirety as a first order approximation. COVID-19 case counts were provided by the SCDHEC for both the ZIP-code level approximation and the actual sewershed delineated for each WWTP. To generate sewershed-level case counts, positive cases within ZIP-codes were further mapped to WWTP boundaries generated from Geographic Information System (GIS) shape files using street-level address encoded data. Supplementary Table S1 gives a comparison of the case counts tabulated within the sewershed approximated by ZIP-codes compared to case counts tabulated within the actual sewershed boundaries. Wastewater collection times and flow characteristics during collection were obtained directly from the operations staff at each WWTP and these data are presented in Supplementary Dataset S1.

Fig. 1.

Wastewater treatment plants in South Carolina participating in the state-wide surveillance of SARS-CoV-2.

Table 1.

Characteristics of WWTPs in this study. MGD=million gallons per day. The population was approximated for each sewershed from ZIP-code-level 2019 U.S. Census estimate. ZIP-codes only partially within the sewershed boundary were counted in their entirety.

WWTP	Initial sample date	Population	Average influent flow rate [MGD]	Thousands served per MGD	Total new cases during study period 11	Total new cases during study period 21
1	4/19/2020	363,714	39.90	9.116	4044	3281
2	4/19/2020	89,624	9.72	9.221	863	753
3	4/19/2020	222,209	10.39	21.387	2060	2272
4	4/19/2020	22,266	1.62	13.744	267	178
5	4/19/2020	264,117	12.19	21.667	2472	2643
6	4/19/2020	191,055	27.81	6.870	2864	1647
7	4/19/2020	89,612	5.214	17.190	1012	786

WWTP	Total new cases per MGD2	Total non-zero days measured2	Median non-zero new cases/100,0002	County level SVI3
1	183.6	116	12	0.6309 M-H
2	166.3	110	10	0.6309 M-H
3	416.9	113	13	0.442 L-M
4	274.7	87	18	0.4959 L-M
5	419.6	108	13	0.3662 L-M
6	162.2	108	14	0.3156 L-M
7	344.9	104	13	0.3156 L-M

¹Time period 1= April 19, 2020-December 13, 2020, time period 2=December 16, 2020–June 28, 2021.

²During entire study period from April 19, 2020, to July 1, 2021.

³Social Vulnerability Index: M-H = moderate to high level of vulnerability, L-M = low to moderate level of vulnerability.

⁴No influent flow meter available. Effluent flow rate listed.

One liter 24 h composite wastewater samples were collected by refrigerated autosamplers twice a week at the influent site of the WWTPs and transported on ice to the laboratory at the University of South Carolina where they were immediately processed. One mL of bovine respiratory syncytial virus (BRSV) vaccine (∼80 million copies/mL) (INFORCE 3®) was added to one liter of wastewater prior to concentration in order to quantify processing and viral extraction efficiency. The samples were then homogenized for 10 min using laboratory blenders and 250 mL of homogenized wastewater was decanted into centrifuge bottles. The procedure for determining the optimal homogenization method has been described in a previous study (Sellers et al., 2022). The samples were centrifuged using an Avanti® J-E Centrifuge (Beckman Coulter Lifesciences, Indianapolis, Indiana) with a JS-5.3 rotor for 30 min at 4577 g and 4 ˚C without braking. The pellets were stored at -80 ˚C and 50 mL of the supernatants were concentrated at 4 ˚C to 400 µL using Millipore Amicon 30 kDa ultrafilters.

2.2. Spatial analysis using ArcGIS

Sewershed maps were created with ArcGIS Pro software (ver. 2.8.7). Shapefiles delineating WWTP sewershed boundaries were obtained directly from each utility. A 2010 national 5-digit zip code boundary TIGER/Line shapefile (updated October 2021) was sourced from the United States Census Bureau and cropped for specific zip codes served by each wastewater treatment plant/sewershed. Locations of wastewater treatment plants were identified with geographic coordinates, converted to the appropriate ArcGIS format and inserted using the X,Y location feature. Supplementary Fig. S1 shows the location of WWTPs and their sewershed boundaries in relation to surrounding zip codes. Social vulnerability maps were created using 2018 South Carolina CDC/ATSDR Social Vulnerability Index (Flanagan et al., 2020) data at the census track level (https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html). Comprehensive SVI maps were created using the RPL_THEMES variable and socioeconomic SVI maps were created using the RPL_THEME1 variable. The RPL_THEMS variable factors in socioeconomic status, household composition and disability, minority status and language, and housing type and transportation. State sewersheds were overlayed on SVI chloropleth maps of each variable. Higher values indicate higher social vulnerability. Supplementary Fig. S2 depicts the census-level social vulnerability indices for each WWTP sewershed.

2.3. RNA extraction

RNA was extracted from 200 µL of the concentrated supernatant using the Qiagen AllPrep PowerViral extraction kit as per the manufacturer's instructions and eluted in 51 µL of RNase-free water. A NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA) was used to assess quality of the extracted nucleic acids, and concentration was measured using a Invitrogen^TM QuBit^TM 2.0 fluorometer (Thermo Fisher Scientific, Waltham, MA). The extracted nucleic acids were stored at -80 ˚C for up to 48 h. SARS-CoV-2 RNA copy numbers of wastewater samples collected before December 16, 2020, were measured by RT-qPCR. SARS-CoV-2 RNA copy numbers of wastewater samples collected on December 16, 2020, or later were quantified by RT-ddPCR. A comparison of RT-qPCR measurements to RT-ddPCR measurements of the same wastewater samples is presented in Supplementary Fig. S3.

2.4. Reverse transcription quantitative PCR (RT-qPCR)

The N2 CDC diagnostic assay was performed using 5 µL of extracted nucleic acids. The nucleocapsid gene target concentration was measured using an Applied Biosystems 7900HT Fast Real-Time PCR System following the manufacturer's instructions. Reverse transcription was performed using the Luna Universal Probe One-Step RT-qPCR kit (New England Biolab, Cat. No. E3006) following manufacturer's instructions. Primers, probes, and the 2019-nCoV_N2 FAM Fluorophore were supplied in the 2019-nCoV RUO Kit (Integrated DNA Technologies, Cat. No. 10006713). A full list of primer and probe sequences is provided in Supplementary Table S2, and reagent concentrations are given Supplementary Table S3. To prepare the positive control, the 2019-nCoV_N_Positive Control (Integrated DNA Technologies, Cat. No. 10006625) was digested using ScaI-HF (New England Biolabs Cat. No. R3122S) according to the manufacturer's protocol. The digested product was cleaned using SPRIselect beads (Beckman Coulter, Cat. No. B23318) according to the manufacturer's directions. Serial dilutions of the digested and cleaned positive control were prepared with nuclease-free water from 100,000 cp/µL to 25 cp/µL. The concentrations were determined by QuBit on the post-restriction enzyme product. The thermal cycling program for the RT-qPCR was as follows: 10 min at 55 ˚C, 1 min at 95 ˚C, 45 cycles of 10 s at 95 ˚C, and 60 s at 60 ˚C. The software program SDS 2.4 was used to analyze the raw data. The coefficient of determination (R²) for all calibration lines was at least 0.9. A representative calibration curve is depicted in Supplementary Fig. S4. The completed Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) table (Bustin et al., 2009) is available in Supplementary Table S4.

2.5. Reverse transcription droplet digital PCR

Details of the reverse transcription droplet digital PCR assays were performed as described previously (Isanovic et al., 2022) and are also available in the Supplementary Methods. A full list of assays and associated primers and probes is available in Supplementary Table S5, and reagent concentrations are given in Supplementary Table S6. The completed digital MIQE table (Huggett et al., 2013) is available in Supplementary Table S7.

2.6. Normalization of SARS-CoV-2 concentration in wastewater to the pepper mild mottle virus

Pepper Mild Mottle Virus (PMMoV) is one of the most abundant RNA viruses in human feces and is commonly used as a fecal indicator in wastewater epidemiology (Kitajima et al., 2018). To ascertain whether normalization of the SARS-CoV-2 RNA concentration to PMMoV would improve the correlation strength between daily case counts and SARS-CoV-2 concentration in wastewater, we performed a linear regression on a subset of the RT-ddPCR data for WWTP 2 with and without normalization (Supplementary Fig. S5). Visual inspection of the two plots indicated less scatter for the data plotted without normalization to PMMoV, resulting in a higher coefficient of determination.

2.7. Limit of blank and limit of detection for RT-ddPCR

Limit of blank (LOB) and limit of detection (LOD) were calculated for the ddPCR as described previously (Isanovic et al., 2022). Details are also available in the Supplementary Methods. The theoretical LOD for the Exact Diagnostics SARS-CoV-2 standard spiked into filtered effluent was 1.3 copies per reaction (520 copies per liter or 2.716 copies per liter after a log10 transformation).

2.8. Data analysis

The number of SARS-CoV-2 genome copies per liter (GC/L) of wastewater was determined through dimensional analysis as described previously (Isanovic et al., 2022). Details are also available in the Supplementary Methods.

A linear regression was performed using the LINEST function in Microsoft Excel for log-transformed SARS-CoV-2 concentrations assessed in wastewater and daily new COVID-19 case counts for the ZIP codes served by each of the seven WWTP. The N2 gene copies were used to estimate the SARS-CoV-2 concentration. SARS-CoV-2 concentrations that were not detected were replaced by the limit of detection for the RT-ddPCR (520 genome copies/L). Daily case counts with a zero value were not included in the analysis. Case counts were iteratively lagged after the wastewater collection day to identify the lag-time resulting in the maximum coefficient of determination reported by the LINEST function. The P value of all correlations was less than 0.001.

The Pearson's correlation coefficient r was calculated using the PEARSON function in Microsoft Excel.

Principal component analysis was conducted using the prcomp package in R Core Team (2013) and visualized using the ggbiplot package. The heatmap and associated hierarchical clustering were performed using pheatmap in R programming language with default parameters.

3. Results and discussion

As part of a state-wide SARS-CoV-2 wastewater surveillance effort, WWTP influent samples were collected biweekly from April 19, 2020, to July 1, 2021. The WWTPs spanned the state of South Carolina (Fig. 1) and included seven sites serving the following counties: Charleston, Kershaw, Lexington, Richland, and York.

3.1. Droplet digital PCR measurements of SARS-CoV-2 RNA concentrations in wastewater correlate strongly with daily COVID-19 case counts

Daily clinical new case data in the ZIP codes served by the WWTPs (provided by the South Carolina Department of Health and Environmental Control) were compared to the SARS-CoV-2 concentration (genome copies/L) measured from 24 h composite influent samples for each WWTP. Table 1 and Supplementary Table S1 summarize characteristics of the WWTPs, including population served, flow, and infection prevalence. Supplementary Fig. S6 illustrates tracking of SARS-CoV-2 copies/L with daily new case counts. To assess the degree of correlation between wastewater and clinical data, we performed a linear regression of the log transformed daily new case counts per 100,000 individuals to the log-transformed SARS-CoV-2 RNA concentrations for sites 1-7 (Fig. 2 ). Since both the daily new case counts and the SARS-CoV-2 genome copy concentration were log transformed, zero values in SARS-CoV-2 concentration were replaced with the limit of detection (520 genome copies/L wastewater for RT-ddPCR measurements) and days with zero values in case counts were excluded from the regression. Samples whose SARS-CoV-2 RNA concentration was measured by RT-qPCR were analyzed separately from samples whose concentration was determined by RT-ddPCR. The linear regression was performed for iteratively lagged daily case counts of the ZIP codes served by the WWTP against the SARS-CoV-2 RNA concentration in WWTP influent. Separate regressions were performed for case counts lagged zero to three days after wastewater collection. The coefficient of determination for the regression for each WWTP and lag time is given in Table 2 . For all sites, the coefficient of determination was greater for the lagged linear regression except WWTP 6. For the time period where RT-ddPCR was used to measure SARS-CoV-2 RNA concentration (December 16, 2020–June 28, 2021), the optimal lag between SARS-CoV-2 RNA concentration and daily new case counts was two days for all sites except WWTP 6. Since the WWTPs differ significantly in their characteristics (Table 1, Supplementary Table S1 and Fig. 2) and the SARS-CoV-2 RNA concentrations were not normalized to a fecal indicator, it is notable that the optimal lag determined from the RT-ddPCR period was consistent across the majority of the sites. The optimal lag between SARS-CoV-2 RNA concentration and daily case counts for the RT-qPCR period (Apr 19, 2020–Dec 13, 2020) varied between zero and three days. Others have reported that wastewater trends precede clinical trends by 3 days (Wurtzer et al., 2022), 4–10 days (Wu et al., 2022), and 15–23 days (Melvin et al., 2021).

Fig. 2.

Linear regression of log-transformed SARS-CoV-2 concentrations assessed in wastewater and daily new COVID-19 case counts for the ZIP codes served by each WWTP.(A) WWTP 1, (B) WWTP 2, (C) WWTP 3, (D) WWTP 4, (E) WWTP 5, (F) WWTP 6, and (G) WWTP 7. Case counts were lagged two days after the measured SARS-CoV-2 concentration. The N2 gene was quantified as an estimate for SARS-CoV-2 concentration. Blue lines represent RT-qPCR measurements (Apr 19, 2020–December 13, 2020) and orange lines represent RT-ddPCR measurements (December 16, 2020–June 28, 2021). The coefficient of determination for each trendline is depicted in an outlined box of the same color.

Table 2.

Coefficient of determination (COD) for the linear regression of log transformed non-zero new case counts against SARS-CoV-2 concentration in genome copies per liter. New case counts were lagged 1-3 days after the wastewater collection date to determine the best fit. The largest COD for each WWTP and time period is bolded.

Apr 19, 2020–Dec 13, 2020
WWTP	0-day lag	1-day lag	2-day lag	3-day lag
1	0.496	0.5699	0.5648	0.5497
2	0.2198	0.0432	0.1478	0.2709
3	0.3296	0.3905	0.3452	0.2726
4	0.0688	0.1754	0.1717	0.0685
5	0.4125	0.4153	0.3755	0.3058
6	0.2558	0.2096	0.2166	0.2991
7	0.3289	0.2702	0.2565	0.2702

Dec 16, 2020–Jun 28, 2021
WWTP	0-day lag	1-day lag	2-day lag	3-day lag
1	0.5694	0.6332	0.7166	0.6645
2	0.6211	0.5808	0.7036	0.6747
3	0.6766	0.7204	0.7338	0.7075
4	0.0463	0.0918	0.5088	0.1852
5	0.5133	0.4425	0.5719	0.3885
6	0.3937	0.2967	0.2611	0.3057
7	0.2854	0.4200	0.4872	0.4200

Pearson's correlation for the SARS-CoV-2 concentration compared to the two-day lagged daily case counts ranged from 0.38 to 0.86 (Table 3 ). These correlations are in close agreement with those reported for Melvin's index and daily case counts (Melvin et al., 2021), which ranged from 0.38 to 0.78. The slopes obtained from RT-ddPCR measurements for all WWTPs except WWTP 5 were within the standard error reported in the literature (0.62 × log10 gene copies per log10 cumulative reported cases with standard error = 0.12) (Medema et al., 2020). Although RT-ddPCR measurements were more sensitive than RT-qPCR measurements (Supplementary Fig. S3), we were unable to compare correlation strengths obtained using these two different methods due to differences in case counts and testing availability during the RT-qPCR and RT-ddPCR time periods.

Table 3.

SARS-CoV-2 RNA concentration (genome copies/L) in wastewater correlates with two-day lagged daily clinical case count (daily new cases per 100,000) for some WWTPs. Both concentration and case counts were log transformed. Dates with zero case counts were excluded from analysis. When SARS-CoV-2 concentration was not detected, a value equal to the limit of detection (520 genome copies/L) was substituted.

Site	Apr 19, 2020–Dec 13, 2020, Pearson's r	Dec 16, 2020–June 28, 2021, Pearson's r
1	0.75	0.85
2	0.38	0.84
3	0.59	0.86
4	0.41	0.71
5	0.61	0.72
6	0.47	0.53
7	0.51	0.70

3.2. Effects of population size and influent flow rate on correlation strength between SARS-CoV-2 RNA concentration in wastewater and two-day lagged daily case counts

The sites included in this study varied in the population served from 22,266 (site 4) to 363,714 (site 1, Table 1). The average influent flow rate ranged from 2 million gallons per day (MGD) to 40 MGD across all sites. Population served and average influent flow rate across all WWTPs is depicted in Fig. 3 . Sites also varied in the amount of total flow permitted from industry (Supplementary Table S1), resulting in different ratios of population served to average influent flow rate. To assess these differences, the ratio of thousands served per average influent flow rate in MGD was calculated (Table 1).

Fig. 3.

Population served and average influent flow rate for the WWTP sites included this study. MGD=million gallons per day. Bars represent population size and X's represent WWTP flow rate.

Out of the seven WWTPs with a 15-month sampling interval, WWTP 6 exhibited the weakest Pearson's correlation during the second study period (Table 3). WWTP 1, which served the largest population, exhibited the highest Pearson's correlation (0.85 for Dec 16, 2020–June 28, 2021), using a two-day lag between wastewater collection and daily case counts. Notably, the ratio of the population served to the average flow rate in million gallons per day at WWTP 6 was the lowest of WWTPs 1-7 (Table 1). Therefore, comparison of the correlation strength between WWTPs 6 and WWTPs 1-5 and 7 suggests that the ratio of population to flow rate may be a useful metric to assess the suitability of WWTPs for SARS-CoV-2 wastewater surveillance. WWTPs with high flow relative to population served are expected to have a weaker correlation due to dilution of SARS-CoV-2 in wastewater below the limit of detection. Consistent with this hypothesis, WWTP 6 exhibited a high proportion of non-detects (Supplementary Table S1), where SARS-CoV-2 concentration was below the limit of detection but two-day lagged case counts were non-zero. We note that other WWTP characteristics not discussed in this study, such as the average sewer pipe length from catchment to plant, may also influence the percentage of non-detects and the correlation strength between COVID-19 case counts and SARS-CoV-2 concentration. For instance, WWTP 7 also exhibited a high proportion of non-detects, despite having a high ratio of thousands served per MGD. Nevertheless, the ratio of population served to total flow can be useful for preliminary assessment in wastewater surveillance efforts.

3.3. Minimum detectable case counts resulting in a measurable SARS-CoV-2 concentration

The linear regressions for each facility between SARS-CoV-2 concentration in wastewater and two-day lagged daily case counts were used to estimate the minimum reported daily new cases counts per 100,000 that are necessary to cause a SARS-CoV-2 concentration above the limit of detection in wastewater during the time period where RT-ddPCR measurements were taken (Table 4 ). One study estimated that between 1 and 100 viral shedders for every two million individuals are necessary to produce a detectable signal (Hart and Halden, 2020), whereas another study estimated a minimum threshold between 2 virus shedders per 10,000 persons and 1 virus shedder per 1000 (Jørgensen et al., 2020). The minimum number of detectable cases in all WWTPs (Table 4) are the same order of magnitude as the upper limit of the study by Hart and Halden (100 shedders in 2 million individuals). However, it should be noted that the true number of shedders will be in excess of those who tested positive since as many as 40% of confirmed cases are asymptomatic (Ma et al., 2021). Some asymptomatic individuals may test due to requirements by employers or schools or for other reasons, but many cases are unreported.

Table 4.

Minimum detectable cases for each WWTP calculated using trendlines based on data from December 16, 2020–June 28, 2021 or measured. Measured minimum cases refer to the lowest number of cases that resulted in a non-zero SARS-CoV-2 concentration. Only trendlines with a coefficient of determination of at least 0.5 were used. The minimum detectable cases were rounded up to whole numbers. A two-day lag was used for the daily case count for both calculated and measured minimum detectable cases.

WWTP	Calculated minimum detectable cases/100,000	Measured minimum detectable cases/100,000	Measured minimum detectable cases
1	2	2	2
2	2	2	1
3	4	3	5
4	11	5	1
5	5	1	1
6	–	2	2
7	–	2	1

3.4. COVID-19 case counts at the ZIP-code level are sufficient for robust correlations with SARS-CoV-2 concentration in wastewater

In wastewater surveillance efforts, it is often more difficult to obtain sewershed-level COVID-19 case count data compared to ZIP-code level case counts. Some wastewater surveillance studies have used sewershed-level data (Haak et al., 2022; Weidhaas et al., 2021), whereas other studies have used county-level or ZIP-code level data (Melvin et al., 2021; Wu et al., 2022). We assessed whether the correlation between case counts and SARS-CoV-2 RNA concentration in wastewater was significantly improved by the use of sewershed-level case counts rather than ZIP-code level case counts. Sewershed-level cumulative case counts were only available on a weekly basis. Therefore, the biweekly measurements of SARS-CoV-2 concentration were averaged for the corresponding week. Supplementary Fig. S1 shows the location of WWTPs and their sewershed boundaries in relation to surrounding zip codes. Supplementary Fig. S7 depicts the correlation of sewershed-level COVID-19 case counts with SARS-CoV-2 RNA concentration in wastewater, whereas Supplementary Fig. S8 illustrates the correlation of ZIP-code-level COVID-19 case counts with SARS-CoV-2 concentration. For WWTP 1, 3-4, and 6 the correlations with RT-ddPCR-measured SARS-CoV-2 concentration that used sewershed-level COVID-19 case counts were higher than those obtained using ZIP-code-level case counts. However, correlations using ZIP-code level data were higher for WWTP 2, 5, and 7. Since COVID-19 cases that are asymptomatic or symptomatic but not reported are not included in the sewershed-level COVID-19 case counts, it is possible that for some communities ZIP-code level data may be closer to the true number of infected individuals that are shedding virus if the ZIP-code may contain a greater number of tested individuals than the sewershed. However, there was no statistically significant difference between the coefficients of determination that were calculated for the correlations using sewershed-level case counts and those using ZIP-code-level case counts (paired two-tailed student's t-test, p-value threshold 0.05). These results indicate that case counts at the ZIP-code level may be sufficient to develop a strong correlation between case counts and SARS-CoV-2 RNA concentrations in WWTP influent.

3.5. Lag time analysis for the correlation of the 7-day moving average of daily case counts to SARS-CoV-2 RNA concentration in wastewater

To assess the effect of smoothing the daily case counts using a 7-day moving average, lag time analysis was repeated for the correlation between averaged daily case counts and SARS-CoV-2 concentration in wastewater. The 7-day moving average was iteratively lagged after the wastewater collection date until a maximum correlation strength was achieved (Supplementary Figs. S9 and S10), which occurred as late as 9 days after wastewater collection. The 7-day moving average improved the correlation strength between COVID-19 case counts and SARS-CoV-2 concentration in wastewater. For example, the Pearson's correlation for WWTP 2 improved from 0.84 to 0.91 by using a 7-day moving average for the daily case counts (Table 3 and Supplementary Figs. S9 and S10).However, the improvements in correlation strength by optimizing the lag time were limited. For instance, optimization of the lag time for WWTP 1, 3, and 7 only resulted in improvements of 1–5% in the Pearson's correlation (Supplementary Fig. S9). Therefore, data smoothing dampened the effect of lag-time optimization. This finding indicates that data smoothing improves correlation strength, but does not significantly affect how far in advance the community will be aware of surges in COVID-19 case counts.

3.6. Community infection rate drives temporal trends in SARS-CoV-2 concentration measured at the WWTP

To quantitatively compare trends in SARS-CoV-2 RNA concentration over time across WWTPs, we first visualized the temporal profiles in a heatmap using R (Fig. 4 ). We used complete linkage hierarchical clustering based on Euclidean distances built into the R function pheatmap to group the WWTPs based on their profiles. The WWTPs broadly fell into two groups: (1) WWTPs 4, 6, and 7, and (2) WWTPs 1-3 and 5. Group 1 was characterized by SARS-CoV-2 concentration measurements below the limit of detection in the early months of the COVID-19 pandemic (April–June 2020), whereas wastewater samples from Group 2 WWTPs contained measurable SARS-CoV-2 RNA concentrations consistently starting on April 29, 2020. Notably, WWTPs 1-3 were located in close proximity in the center of South Carolina (Fig. 1), suggesting that transmission dynamics from these communities may have influenced each other. Group 2 WWTPs all served populations greater than 100,000 individuals, except for WWTP 2. Conversely, Group 1 WWTPs served less than 100,000 individuals, except for WWTP 6. Within Group 2, WWTPs 2 and 3 showed similar SARS-CoV-2 RNA concentration profiles over time, despite the fact that the sewershed for WWTP 3 included 132,585 more individuals than WWTP 2. The average flow of WWTPs 2 and 3 were similar (Fig. 3 and Table 1). The total number of infected individuals in areas served by WWTPs 2 and 3 from April 19 to December 13, 2020, were 863 and 2060, respectively, representing infection rates of 963 and 927 cases/100,000. WWTP 5, which was also included in Group 2, had an infection rate of 935 during the same time period. These results suggest that one of the most important factors driving the temporal profile of SARS-CoV-2 concentration in sewage is the relative number of infected individuals, irrespective of the average flow rate or total population served.

Fig. 4.

Logarithm of SARS-CoV-2 copies/L measured by the N2 gene assay in wastewater over time. Sampling dates are arranged in descending order from 4/19/2020 (top row) to 12/13/2020 (bottom row). Cells shown as zero indicate that SARS-CoV-2 was not detected. Only dates were included when all seven sites were sampled synchronously.

3.7. Principal component analysis of WWTP characteristics

To assess how the different characteristics of each WWTP might be related to the temporal trends, we performed a principal component analysis (PCA) using the data from Tables 1 and 3 (RT-ddPCR time period). PCA has been used in other wastewater-based epidemiology studies to assess the association between PMMoV threshold cycle and site-specific influent flow rate variables (Melvin et al., 2021), to better understand the relatedness between treatment process parameters and the removal of SARS-CoV-2 gene targets (Kumar et al., 2021), and to examine the similarity in SARS-CoV-2 RNA profiles over time across spatial sampling sites (Haak et al., 2022). In the present work, the importance of each PC in explaining the variance in correlation strengths and WWTP characteristics, as well as the PCA loadings and scores are presented in Supplementary Tables S8–S10. PC1 explained 63.1% of the variance and PC2 and 3 explained 14.9% and 12.8%, respectively. A plot of the position of WWTPs 1-7 relative to PC1 and PC2 is depicted in Fig. 5 . WWTP 3, 5, and 7 are grouped at higher values of PC2, whereas WWTPs 1-2, 4, and 6 were distinct and separated along PC1 and PC2. Visualization of the variable axes against the principal components indicated that the largest contribution to PC2 was from the ratio of population served to average flow rate (thousands served per MGD). Social vulnerability index was anticorrelated with PC2. The majority of the remaining variables were anticorrelated with PC1, except for the median non-zero cases/100,000 during study period and total new cases/100,000 during study period. The results of the PCA analysis suggest that the communities served by WWTPs 3, 5, and 7 were similar in social vulnerability and ratio of population served to flow rate. Likewise, the correlation strength, as assessed by the Pearson's r, was similar for WWTPs 3, 5, and 7.

Fig. 5.

Principal component analysis (PCA) of WWTP characteristics and correlation strengths. PCA was performed using the prcomp function and visualized using ggbiplot in the R programming language. The input matrix to the PCA was Table 1, excluding the Initial sample date column, concatenated with Table 3. Importance of principal components, loadings, and scores are presenting in Supplementary Tables S4–S6. Variable axes are as follows: Pop=population, Flow=average influent flow rate [MGD], P/F=thousands served per MGD, C1=total new cases from April 19, 2020 to December 13, 2020, C2=total new cases from December 16, 2020 to June 28, 2021, C_norm=total new cases/100,000 per MGD during both study periods, NZ=total non-zero days, C_med=median non-zero new cases/100,000 during both study periods, r1=Pearson's r for April 19, 2020-December 13, 2020, r2=Pearson's r for December 16, 2020-June 28, 2021, and SVI=social vulnerability index.

4. Conclusion

This study represented one of the most comprehensive wastewater surveillance efforts in South Carolina, spanning five counties across the state and a period of 15 months. A consistent lag time of two days was observed between clinical case counts and SARS_CoV-2 concentration for six of the seven WWTPs, despite differences in average flow rate and size of population served, suggesting that for most WWTPs a forecasting window of at least two days is attainable. The SARS-CoV-2 concentration profiles over time from April to mid-December 2020 exhibited similarity for WWTPs with similar infection rates (total cases per 100,000 individuals). Comparison of the Pearson's correlation between the SARS-CoV-2 concentration in wastewater and the daily COVID-19 case count suggested that a strong correlation (Pearson's correlation > 0.7) requires a minimum ratio of population served to average flow rate. As the National Wastewater Surveillance System continues to expand, more statistical power provided by data from additional sites will improve our understanding of the major factors that impact the correlation strength and lag between SARS-CoV-2 concentration in wastewater and clinical case counts. Notably, a seven-day moving average of daily case counts improved the correlation between case counts and SARS-CoV-2 concentration in wastewater and dampened the effect of lag-time optimization. We also note that case count data at the ZIP-code level was sufficient for robust correlations in this study, although sewershed-level case counts improved correlation strengths for some WWTPs.

CRediT authorship contribution statement

Candice L. Swift: Writing – original draft, Formal analysis, Investigation. Mirza Isanovic: Investigation, Formal analysis, Writing – review & editing. Karlen E. Correa Velez: Investigation, Writing – review & editing. R. Sean Norman: Conceptualization, Investigation, Writing – review & editing, Funding acquisition, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

• Data will be made available on request.